Cochlear modeling using a general purpose digital signal processor

An effort to design a computationally modified cochlear model and implement such a model in a digital signal processor system is described. The model is based on the analog cochlear filter and on Seneff's cochlear model. These two models were evaluated and combined. Two functional blocks are added to the analog filter bank to improve the frequency response. A 20-channel filter bank was constructed covering the frequency spectrum from 200 Hz to 3500 Hz. The hair cell section defined by Seneff was refined and incorporated into the model. A new design for the Generalized Synchrony Detector of the Seneff model is presented. The modified cochlear model was extensively tested with various input stimuli, including speech signals. This simulation demonstrated the proper operation of the new model including dynamic range compression, adaptation, and two-tone suppression. The modified cochlear model is implemented on a Motorola DSP56000 signal processing system. The current implementation simulates one channel of the cochlear model in real time.<
>